Root/drivers/md/persistent-data/dm-btree.c

1/*
2 * Copyright (C) 2011 Red Hat, Inc.
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-btree-internal.h"
8#include "dm-space-map.h"
9#include "dm-transaction-manager.h"
10
11#include <linux/export.h>
12#include <linux/device-mapper.h>
13
14#define DM_MSG_PREFIX "btree"
15
16/*----------------------------------------------------------------
17 * Array manipulation
18 *--------------------------------------------------------------*/
19static void memcpy_disk(void *dest, const void *src, size_t len)
20    __dm_written_to_disk(src)
21{
22    memcpy(dest, src, len);
23    __dm_unbless_for_disk(src);
24}
25
26static void array_insert(void *base, size_t elt_size, unsigned nr_elts,
27             unsigned index, void *elt)
28    __dm_written_to_disk(elt)
29{
30    if (index < nr_elts)
31        memmove(base + (elt_size * (index + 1)),
32            base + (elt_size * index),
33            (nr_elts - index) * elt_size);
34
35    memcpy_disk(base + (elt_size * index), elt, elt_size);
36}
37
38/*----------------------------------------------------------------*/
39
40/* makes the assumption that no two keys are the same. */
41static int bsearch(struct node *n, uint64_t key, int want_hi)
42{
43    int lo = -1, hi = le32_to_cpu(n->header.nr_entries);
44
45    while (hi - lo > 1) {
46        int mid = lo + ((hi - lo) / 2);
47        uint64_t mid_key = le64_to_cpu(n->keys[mid]);
48
49        if (mid_key == key)
50            return mid;
51
52        if (mid_key < key)
53            lo = mid;
54        else
55            hi = mid;
56    }
57
58    return want_hi ? hi : lo;
59}
60
61int lower_bound(struct node *n, uint64_t key)
62{
63    return bsearch(n, key, 0);
64}
65
66void inc_children(struct dm_transaction_manager *tm, struct node *n,
67          struct dm_btree_value_type *vt)
68{
69    unsigned i;
70    uint32_t nr_entries = le32_to_cpu(n->header.nr_entries);
71
72    if (le32_to_cpu(n->header.flags) & INTERNAL_NODE)
73        for (i = 0; i < nr_entries; i++)
74            dm_tm_inc(tm, value64(n, i));
75    else if (vt->inc)
76        for (i = 0; i < nr_entries; i++)
77            vt->inc(vt->context, value_ptr(n, i));
78}
79
80static int insert_at(size_t value_size, struct node *node, unsigned index,
81              uint64_t key, void *value)
82              __dm_written_to_disk(value)
83{
84    uint32_t nr_entries = le32_to_cpu(node->header.nr_entries);
85    __le64 key_le = cpu_to_le64(key);
86
87    if (index > nr_entries ||
88        index >= le32_to_cpu(node->header.max_entries)) {
89        DMERR("too many entries in btree node for insert");
90        __dm_unbless_for_disk(value);
91        return -ENOMEM;
92    }
93
94    __dm_bless_for_disk(&key_le);
95
96    array_insert(node->keys, sizeof(*node->keys), nr_entries, index, &key_le);
97    array_insert(value_base(node), value_size, nr_entries, index, value);
98    node->header.nr_entries = cpu_to_le32(nr_entries + 1);
99
100    return 0;
101}
102
103/*----------------------------------------------------------------*/
104
105/*
106 * We want 3n entries (for some n). This works more nicely for repeated
107 * insert remove loops than (2n + 1).
108 */
109static uint32_t calc_max_entries(size_t value_size, size_t block_size)
110{
111    uint32_t total, n;
112    size_t elt_size = sizeof(uint64_t) + value_size; /* key + value */
113
114    block_size -= sizeof(struct node_header);
115    total = block_size / elt_size;
116    n = total / 3; /* rounds down */
117
118    return 3 * n;
119}
120
121int dm_btree_empty(struct dm_btree_info *info, dm_block_t *root)
122{
123    int r;
124    struct dm_block *b;
125    struct node *n;
126    size_t block_size;
127    uint32_t max_entries;
128
129    r = new_block(info, &b);
130    if (r < 0)
131        return r;
132
133    block_size = dm_bm_block_size(dm_tm_get_bm(info->tm));
134    max_entries = calc_max_entries(info->value_type.size, block_size);
135
136    n = dm_block_data(b);
137    memset(n, 0, block_size);
138    n->header.flags = cpu_to_le32(LEAF_NODE);
139    n->header.nr_entries = cpu_to_le32(0);
140    n->header.max_entries = cpu_to_le32(max_entries);
141    n->header.value_size = cpu_to_le32(info->value_type.size);
142
143    *root = dm_block_location(b);
144    return unlock_block(info, b);
145}
146EXPORT_SYMBOL_GPL(dm_btree_empty);
147
148/*----------------------------------------------------------------*/
149
150/*
151 * Deletion uses a recursive algorithm, since we have limited stack space
152 * we explicitly manage our own stack on the heap.
153 */
154#define MAX_SPINE_DEPTH 64
155struct frame {
156    struct dm_block *b;
157    struct node *n;
158    unsigned level;
159    unsigned nr_children;
160    unsigned current_child;
161};
162
163struct del_stack {
164    struct dm_transaction_manager *tm;
165    int top;
166    struct frame spine[MAX_SPINE_DEPTH];
167};
168
169static int top_frame(struct del_stack *s, struct frame **f)
170{
171    if (s->top < 0) {
172        DMERR("btree deletion stack empty");
173        return -EINVAL;
174    }
175
176    *f = s->spine + s->top;
177
178    return 0;
179}
180
181static int unprocessed_frames(struct del_stack *s)
182{
183    return s->top >= 0;
184}
185
186static int push_frame(struct del_stack *s, dm_block_t b, unsigned level)
187{
188    int r;
189    uint32_t ref_count;
190
191    if (s->top >= MAX_SPINE_DEPTH - 1) {
192        DMERR("btree deletion stack out of memory");
193        return -ENOMEM;
194    }
195
196    r = dm_tm_ref(s->tm, b, &ref_count);
197    if (r)
198        return r;
199
200    if (ref_count > 1)
201        /*
202         * This is a shared node, so we can just decrement it's
203         * reference counter and leave the children.
204         */
205        dm_tm_dec(s->tm, b);
206
207    else {
208        struct frame *f = s->spine + ++s->top;
209
210        r = dm_tm_read_lock(s->tm, b, &btree_node_validator, &f->b);
211        if (r) {
212            s->top--;
213            return r;
214        }
215
216        f->n = dm_block_data(f->b);
217        f->level = level;
218        f->nr_children = le32_to_cpu(f->n->header.nr_entries);
219        f->current_child = 0;
220    }
221
222    return 0;
223}
224
225static void pop_frame(struct del_stack *s)
226{
227    struct frame *f = s->spine + s->top--;
228
229    dm_tm_dec(s->tm, dm_block_location(f->b));
230    dm_tm_unlock(s->tm, f->b);
231}
232
233int dm_btree_del(struct dm_btree_info *info, dm_block_t root)
234{
235    int r;
236    struct del_stack *s;
237
238    s = kmalloc(sizeof(*s), GFP_KERNEL);
239    if (!s)
240        return -ENOMEM;
241    s->tm = info->tm;
242    s->top = -1;
243
244    r = push_frame(s, root, 1);
245    if (r)
246        goto out;
247
248    while (unprocessed_frames(s)) {
249        uint32_t flags;
250        struct frame *f;
251        dm_block_t b;
252
253        r = top_frame(s, &f);
254        if (r)
255            goto out;
256
257        if (f->current_child >= f->nr_children) {
258            pop_frame(s);
259            continue;
260        }
261
262        flags = le32_to_cpu(f->n->header.flags);
263        if (flags & INTERNAL_NODE) {
264            b = value64(f->n, f->current_child);
265            f->current_child++;
266            r = push_frame(s, b, f->level);
267            if (r)
268                goto out;
269
270        } else if (f->level != (info->levels - 1)) {
271            b = value64(f->n, f->current_child);
272            f->current_child++;
273            r = push_frame(s, b, f->level + 1);
274            if (r)
275                goto out;
276
277        } else {
278            if (info->value_type.dec) {
279                unsigned i;
280
281                for (i = 0; i < f->nr_children; i++)
282                    info->value_type.dec(info->value_type.context,
283                                 value_ptr(f->n, i));
284            }
285            f->current_child = f->nr_children;
286        }
287    }
288
289out:
290    kfree(s);
291    return r;
292}
293EXPORT_SYMBOL_GPL(dm_btree_del);
294
295/*----------------------------------------------------------------*/
296
297static int btree_lookup_raw(struct ro_spine *s, dm_block_t block, uint64_t key,
298                int (*search_fn)(struct node *, uint64_t),
299                uint64_t *result_key, void *v, size_t value_size)
300{
301    int i, r;
302    uint32_t flags, nr_entries;
303
304    do {
305        r = ro_step(s, block);
306        if (r < 0)
307            return r;
308
309        i = search_fn(ro_node(s), key);
310
311        flags = le32_to_cpu(ro_node(s)->header.flags);
312        nr_entries = le32_to_cpu(ro_node(s)->header.nr_entries);
313        if (i < 0 || i >= nr_entries)
314            return -ENODATA;
315
316        if (flags & INTERNAL_NODE)
317            block = value64(ro_node(s), i);
318
319    } while (!(flags & LEAF_NODE));
320
321    *result_key = le64_to_cpu(ro_node(s)->keys[i]);
322    memcpy(v, value_ptr(ro_node(s), i), value_size);
323
324    return 0;
325}
326
327int dm_btree_lookup(struct dm_btree_info *info, dm_block_t root,
328            uint64_t *keys, void *value_le)
329{
330    unsigned level, last_level = info->levels - 1;
331    int r = -ENODATA;
332    uint64_t rkey;
333    __le64 internal_value_le;
334    struct ro_spine spine;
335
336    init_ro_spine(&spine, info);
337    for (level = 0; level < info->levels; level++) {
338        size_t size;
339        void *value_p;
340
341        if (level == last_level) {
342            value_p = value_le;
343            size = info->value_type.size;
344
345        } else {
346            value_p = &internal_value_le;
347            size = sizeof(uint64_t);
348        }
349
350        r = btree_lookup_raw(&spine, root, keys[level],
351                     lower_bound, &rkey,
352                     value_p, size);
353
354        if (!r) {
355            if (rkey != keys[level]) {
356                exit_ro_spine(&spine);
357                return -ENODATA;
358            }
359        } else {
360            exit_ro_spine(&spine);
361            return r;
362        }
363
364        root = le64_to_cpu(internal_value_le);
365    }
366    exit_ro_spine(&spine);
367
368    return r;
369}
370EXPORT_SYMBOL_GPL(dm_btree_lookup);
371
372/*
373 * Splits a node by creating a sibling node and shifting half the nodes
374 * contents across. Assumes there is a parent node, and it has room for
375 * another child.
376 *
377 * Before:
378 * +--------+
379 * | Parent |
380 * +--------+
381 * |
382 * v
383 * +----------+
384 * | A ++++++ |
385 * +----------+
386 *
387 *
388 * After:
389 * +--------+
390 * | Parent |
391 * +--------+
392 * | |
393 * v +------+
394 * +---------+ |
395 * | A* +++ | v
396 * +---------+ +-------+
397 * | B +++ |
398 * +-------+
399 *
400 * Where A* is a shadow of A.
401 */
402static int btree_split_sibling(struct shadow_spine *s, dm_block_t root,
403                   unsigned parent_index, uint64_t key)
404{
405    int r;
406    size_t size;
407    unsigned nr_left, nr_right;
408    struct dm_block *left, *right, *parent;
409    struct node *ln, *rn, *pn;
410    __le64 location;
411
412    left = shadow_current(s);
413
414    r = new_block(s->info, &right);
415    if (r < 0)
416        return r;
417
418    ln = dm_block_data(left);
419    rn = dm_block_data(right);
420
421    nr_left = le32_to_cpu(ln->header.nr_entries) / 2;
422    nr_right = le32_to_cpu(ln->header.nr_entries) - nr_left;
423
424    ln->header.nr_entries = cpu_to_le32(nr_left);
425
426    rn->header.flags = ln->header.flags;
427    rn->header.nr_entries = cpu_to_le32(nr_right);
428    rn->header.max_entries = ln->header.max_entries;
429    rn->header.value_size = ln->header.value_size;
430    memcpy(rn->keys, ln->keys + nr_left, nr_right * sizeof(rn->keys[0]));
431
432    size = le32_to_cpu(ln->header.flags) & INTERNAL_NODE ?
433        sizeof(uint64_t) : s->info->value_type.size;
434    memcpy(value_ptr(rn, 0), value_ptr(ln, nr_left),
435           size * nr_right);
436
437    /*
438     * Patch up the parent
439     */
440    parent = shadow_parent(s);
441
442    pn = dm_block_data(parent);
443    location = cpu_to_le64(dm_block_location(left));
444    __dm_bless_for_disk(&location);
445    memcpy_disk(value_ptr(pn, parent_index),
446            &location, sizeof(__le64));
447
448    location = cpu_to_le64(dm_block_location(right));
449    __dm_bless_for_disk(&location);
450
451    r = insert_at(sizeof(__le64), pn, parent_index + 1,
452              le64_to_cpu(rn->keys[0]), &location);
453    if (r)
454        return r;
455
456    if (key < le64_to_cpu(rn->keys[0])) {
457        unlock_block(s->info, right);
458        s->nodes[1] = left;
459    } else {
460        unlock_block(s->info, left);
461        s->nodes[1] = right;
462    }
463
464    return 0;
465}
466
467/*
468 * Splits a node by creating two new children beneath the given node.
469 *
470 * Before:
471 * +----------+
472 * | A ++++++ |
473 * +----------+
474 *
475 *
476 * After:
477 * +------------+
478 * | A (shadow) |
479 * +------------+
480 * | |
481 * +------+ +----+
482 * | |
483 * v v
484 * +-------+ +-------+
485 * | B +++ | | C +++ |
486 * +-------+ +-------+
487 */
488static int btree_split_beneath(struct shadow_spine *s, uint64_t key)
489{
490    int r;
491    size_t size;
492    unsigned nr_left, nr_right;
493    struct dm_block *left, *right, *new_parent;
494    struct node *pn, *ln, *rn;
495    __le64 val;
496
497    new_parent = shadow_current(s);
498
499    r = new_block(s->info, &left);
500    if (r < 0)
501        return r;
502
503    r = new_block(s->info, &right);
504    if (r < 0) {
505        /* FIXME: put left */
506        return r;
507    }
508
509    pn = dm_block_data(new_parent);
510    ln = dm_block_data(left);
511    rn = dm_block_data(right);
512
513    nr_left = le32_to_cpu(pn->header.nr_entries) / 2;
514    nr_right = le32_to_cpu(pn->header.nr_entries) - nr_left;
515
516    ln->header.flags = pn->header.flags;
517    ln->header.nr_entries = cpu_to_le32(nr_left);
518    ln->header.max_entries = pn->header.max_entries;
519    ln->header.value_size = pn->header.value_size;
520
521    rn->header.flags = pn->header.flags;
522    rn->header.nr_entries = cpu_to_le32(nr_right);
523    rn->header.max_entries = pn->header.max_entries;
524    rn->header.value_size = pn->header.value_size;
525
526    memcpy(ln->keys, pn->keys, nr_left * sizeof(pn->keys[0]));
527    memcpy(rn->keys, pn->keys + nr_left, nr_right * sizeof(pn->keys[0]));
528
529    size = le32_to_cpu(pn->header.flags) & INTERNAL_NODE ?
530        sizeof(__le64) : s->info->value_type.size;
531    memcpy(value_ptr(ln, 0), value_ptr(pn, 0), nr_left * size);
532    memcpy(value_ptr(rn, 0), value_ptr(pn, nr_left),
533           nr_right * size);
534
535    /* new_parent should just point to l and r now */
536    pn->header.flags = cpu_to_le32(INTERNAL_NODE);
537    pn->header.nr_entries = cpu_to_le32(2);
538    pn->header.max_entries = cpu_to_le32(
539        calc_max_entries(sizeof(__le64),
540                 dm_bm_block_size(
541                     dm_tm_get_bm(s->info->tm))));
542    pn->header.value_size = cpu_to_le32(sizeof(__le64));
543
544    val = cpu_to_le64(dm_block_location(left));
545    __dm_bless_for_disk(&val);
546    pn->keys[0] = ln->keys[0];
547    memcpy_disk(value_ptr(pn, 0), &val, sizeof(__le64));
548
549    val = cpu_to_le64(dm_block_location(right));
550    __dm_bless_for_disk(&val);
551    pn->keys[1] = rn->keys[0];
552    memcpy_disk(value_ptr(pn, 1), &val, sizeof(__le64));
553
554    /*
555     * rejig the spine. This is ugly, since it knows too
556     * much about the spine
557     */
558    if (s->nodes[0] != new_parent) {
559        unlock_block(s->info, s->nodes[0]);
560        s->nodes[0] = new_parent;
561    }
562    if (key < le64_to_cpu(rn->keys[0])) {
563        unlock_block(s->info, right);
564        s->nodes[1] = left;
565    } else {
566        unlock_block(s->info, left);
567        s->nodes[1] = right;
568    }
569    s->count = 2;
570
571    return 0;
572}
573
574static int btree_insert_raw(struct shadow_spine *s, dm_block_t root,
575                struct dm_btree_value_type *vt,
576                uint64_t key, unsigned *index)
577{
578    int r, i = *index, top = 1;
579    struct node *node;
580
581    for (;;) {
582        r = shadow_step(s, root, vt);
583        if (r < 0)
584            return r;
585
586        node = dm_block_data(shadow_current(s));
587
588        /*
589         * We have to patch up the parent node, ugly, but I don't
590         * see a way to do this automatically as part of the spine
591         * op.
592         */
593        if (shadow_has_parent(s) && i >= 0) { /* FIXME: second clause unness. */
594            __le64 location = cpu_to_le64(dm_block_location(shadow_current(s)));
595
596            __dm_bless_for_disk(&location);
597            memcpy_disk(value_ptr(dm_block_data(shadow_parent(s)), i),
598                    &location, sizeof(__le64));
599        }
600
601        node = dm_block_data(shadow_current(s));
602
603        if (node->header.nr_entries == node->header.max_entries) {
604            if (top)
605                r = btree_split_beneath(s, key);
606            else
607                r = btree_split_sibling(s, root, i, key);
608
609            if (r < 0)
610                return r;
611        }
612
613        node = dm_block_data(shadow_current(s));
614
615        i = lower_bound(node, key);
616
617        if (le32_to_cpu(node->header.flags) & LEAF_NODE)
618            break;
619
620        if (i < 0) {
621            /* change the bounds on the lowest key */
622            node->keys[0] = cpu_to_le64(key);
623            i = 0;
624        }
625
626        root = value64(node, i);
627        top = 0;
628    }
629
630    if (i < 0 || le64_to_cpu(node->keys[i]) != key)
631        i++;
632
633    *index = i;
634    return 0;
635}
636
637static int insert(struct dm_btree_info *info, dm_block_t root,
638          uint64_t *keys, void *value, dm_block_t *new_root,
639          int *inserted)
640          __dm_written_to_disk(value)
641{
642    int r, need_insert;
643    unsigned level, index = -1, last_level = info->levels - 1;
644    dm_block_t block = root;
645    struct shadow_spine spine;
646    struct node *n;
647    struct dm_btree_value_type le64_type;
648
649    le64_type.context = NULL;
650    le64_type.size = sizeof(__le64);
651    le64_type.inc = NULL;
652    le64_type.dec = NULL;
653    le64_type.equal = NULL;
654
655    init_shadow_spine(&spine, info);
656
657    for (level = 0; level < (info->levels - 1); level++) {
658        r = btree_insert_raw(&spine, block, &le64_type, keys[level], &index);
659        if (r < 0)
660            goto bad;
661
662        n = dm_block_data(shadow_current(&spine));
663        need_insert = ((index >= le32_to_cpu(n->header.nr_entries)) ||
664                   (le64_to_cpu(n->keys[index]) != keys[level]));
665
666        if (need_insert) {
667            dm_block_t new_tree;
668            __le64 new_le;
669
670            r = dm_btree_empty(info, &new_tree);
671            if (r < 0)
672                goto bad;
673
674            new_le = cpu_to_le64(new_tree);
675            __dm_bless_for_disk(&new_le);
676
677            r = insert_at(sizeof(uint64_t), n, index,
678                      keys[level], &new_le);
679            if (r)
680                goto bad;
681        }
682
683        if (level < last_level)
684            block = value64(n, index);
685    }
686
687    r = btree_insert_raw(&spine, block, &info->value_type,
688                 keys[level], &index);
689    if (r < 0)
690        goto bad;
691
692    n = dm_block_data(shadow_current(&spine));
693    need_insert = ((index >= le32_to_cpu(n->header.nr_entries)) ||
694               (le64_to_cpu(n->keys[index]) != keys[level]));
695
696    if (need_insert) {
697        if (inserted)
698            *inserted = 1;
699
700        r = insert_at(info->value_type.size, n, index,
701                  keys[level], value);
702        if (r)
703            goto bad_unblessed;
704    } else {
705        if (inserted)
706            *inserted = 0;
707
708        if (info->value_type.dec &&
709            (!info->value_type.equal ||
710             !info->value_type.equal(
711                 info->value_type.context,
712                 value_ptr(n, index),
713                 value))) {
714            info->value_type.dec(info->value_type.context,
715                         value_ptr(n, index));
716        }
717        memcpy_disk(value_ptr(n, index),
718                value, info->value_type.size);
719    }
720
721    *new_root = shadow_root(&spine);
722    exit_shadow_spine(&spine);
723
724    return 0;
725
726bad:
727    __dm_unbless_for_disk(value);
728bad_unblessed:
729    exit_shadow_spine(&spine);
730    return r;
731}
732
733int dm_btree_insert(struct dm_btree_info *info, dm_block_t root,
734            uint64_t *keys, void *value, dm_block_t *new_root)
735            __dm_written_to_disk(value)
736{
737    return insert(info, root, keys, value, new_root, NULL);
738}
739EXPORT_SYMBOL_GPL(dm_btree_insert);
740
741int dm_btree_insert_notify(struct dm_btree_info *info, dm_block_t root,
742               uint64_t *keys, void *value, dm_block_t *new_root,
743               int *inserted)
744               __dm_written_to_disk(value)
745{
746    return insert(info, root, keys, value, new_root, inserted);
747}
748EXPORT_SYMBOL_GPL(dm_btree_insert_notify);
749
750/*----------------------------------------------------------------*/
751
752static int find_highest_key(struct ro_spine *s, dm_block_t block,
753                uint64_t *result_key, dm_block_t *next_block)
754{
755    int i, r;
756    uint32_t flags;
757
758    do {
759        r = ro_step(s, block);
760        if (r < 0)
761            return r;
762
763        flags = le32_to_cpu(ro_node(s)->header.flags);
764        i = le32_to_cpu(ro_node(s)->header.nr_entries);
765        if (!i)
766            return -ENODATA;
767        else
768            i--;
769
770        *result_key = le64_to_cpu(ro_node(s)->keys[i]);
771        if (next_block || flags & INTERNAL_NODE)
772            block = value64(ro_node(s), i);
773
774    } while (flags & INTERNAL_NODE);
775
776    if (next_block)
777        *next_block = block;
778    return 0;
779}
780
781int dm_btree_find_highest_key(struct dm_btree_info *info, dm_block_t root,
782                  uint64_t *result_keys)
783{
784    int r = 0, count = 0, level;
785    struct ro_spine spine;
786
787    init_ro_spine(&spine, info);
788    for (level = 0; level < info->levels; level++) {
789        r = find_highest_key(&spine, root, result_keys + level,
790                     level == info->levels - 1 ? NULL : &root);
791        if (r == -ENODATA) {
792            r = 0;
793            break;
794
795        } else if (r)
796            break;
797
798        count++;
799    }
800    exit_ro_spine(&spine);
801
802    return r ? r : count;
803}
804EXPORT_SYMBOL_GPL(dm_btree_find_highest_key);
805

Archive Download this file



interactive